The general aim of this proposal is to investigate how people produce and comprehend speech given the real-time constraints under which they operate. We will: (1) analyse a large corpus of speech error data so as to gain information on the representation of words, the organization of the lexicon, the hierarchy of units and levels, the serial ordering of events in speech production, and the kinds of syntactic (e.g. transformational) rules which may operate in the production of utterances; and (2) conduct a number of experiments in order to gain further information concerning units, rules, and stages in the production and comprehension of speech. All of the research is aimed at acquiring more data of the kind that can lead to a viable model of linguistic performance. We wish to see if and how the concepts of contemporary linguistics are appropriate in the construction of such a model.